Automatic Induction of a POS Tagset for Italian

Proceedings of the Australasian Language Technology Workshop 2005, pages 176–183, Sydney, Australia, December 2005. Automatic induction of a POS tagset for Italian R. Bernardi KRDB, Free University of Bolzano Bozen, P.zza Domenicani, 3 39100 Bolzano Bozen, Italy, bernardi@inf.unibz.it A. Bolognesi, C. Seidenari and F. Tamburini CILTA, University of Bologna, P.zza San Giovanni in Monte, 4, I-40124, Bologna, Italy, {bolognesi,seidenari,tamburini}@cilta.unibo.it Abstract In this paper we present work in progress on the PoS annotation of an Italian Corpus (CORIS) developed at CILTA (University of Bologna). We aim to automatically induce the PoS tagset by analysing the distributional behaviour of Italian words by relying only on theory-neutral linguistic knowledge. To this end, we propose an algorithm that derives a possible tagset to be further interpreted and defined by the linguist. The algorithm extracts information from loosely labelled dependency structures that encode only basic and broadly accepted syntactic relations, namely Head/Dependent, and the distinction of dependents into Argument vs. Adjunct. 1 TheIntroduction work presented in this paper is part of a project aiming to annotate CORIS/CODIS (Rossini Favretti et al., 2002), a 100-million-word synchronic corpus of con- temporary written Italian, with part-of-speech (PoS) tags. Italian is one of the languages for which a set of annotation guidelines has been developed in the context of the EAGLES project (Monachini, 1995). Several research groups have worked on PoS annotation in practice (for example, Torino University, Xerox and Venice University), but comparing the tag sets used by these groups with Monachini’s guidelines reveals that though there is a general agreement on the main parts of speech to be used1, considerable divergence exists when it comes to the actual classifica- tion of Italian words with respect to these main PoS classes. The classes for which differences of opinion are most evident are adjectives, de- terminers and adverbs. For instance, words like 1The standard classification consists of nouns, verbs, adjectives, determiners, articles, adverbs, prepositions, conjunctions, numerals, interjections, punctuation and a class of residual items which differs from project to project. molti (many) have been classified as “indefinite determiners” by Monachini, “plural quantifiers” by Xerox, “indefinite adjectives” by the Venice and Turin groups. It is not simply a matter of different terminological options resolvable by a mere one-to-one relabelling, nor a matter of sim- ply mapping different classes into a greater one. Crossings between tagsets are complex mostly because of the different theoretical points of view used in categorizing words. For instance, the single tag DET “determiner” in the Xerox tagset matches with DIM “demonstrative ad- jective” or ART “article” in the Venice group (and with DET “determiner” or ART “article” in Monachini) whereas, viceversa, the single tag DEIT “deictic pronoun” by the Venice group matches alternatively with DEM “demonstra- tive” or PRON “personal pronoun” in Xerox. These simple examples show that the choice of PoS tag is already influenced by the underly- ing linguistic theory adopted. This theoretical bias will then influence the kind of conclusions one can draw from the annotated corpus. Our aim is to automatically derive …

Content truncated to 500 words